On today’s episode of TrueNAS Tech Talk, Chris and Kris talk about the first few days of the Fangtooth 25.04 update and the exciting new functionality coming to TrueNAS in this version, provide some clarification about the upcoming Apps change to allow individual IP addresses, and provide the inside scoop on some major features in development for the 25.10 release of Goldeye. Plus, the added benefits of listening to T3 include new career opportunities? Tune in to find out more!
Watched this episode recently, really curious about one of the long term goals. NVME over ethernet (NVME over Fabric). Is it like iscsi? What are some benefits?
Much as iSCSI lets a remote computer (the “initiator”) see a target disk as if it was a locally attached SCSI/SAS device, NVMe/TCP or NVMe-oF will do the same, but the remote computer will think it’s speaking to a locally attached NVMe device.
NVMe has, in general, lower protocol overhead vs. SCSI/SAS, as well as having higher queue depths, the ability to operate in larger segments of data, and should overall provide both better peak throughput and lower latency.
This of course relies on having a fast interlink between your initiatior and target - NVMe/TCP isn’t going to magically make a 1GbE line as fast as locally-attached NVMe, but it will probably start to see measurable benefits above 10GbE, so as the enterprise starts to pick up the beyond-100GbE standards (anyone for 400GbE? 
) they’ll download more 40GbE gear to consumer-attainable levels.
100 GbE NICs may already be found at consumer-attainable level. But the number of PCIe lanes that they require is not consumer-friendly.
I think 10Ge is a good starting point. My Core box will fill a gig or 2.5g connection but not a 10g connection over smb. So I see room for improvement there, and yeah, 10g between the target and the client. 100G is still in slow-your-roll territory and like etorix said, we’re getting cheaped on pcie lanes.
